#Verifiable Computation Yesterday, I heard about an AI business model that made me feel that the potential of verifiable models like ZKML/ZKGPT has yet to be fully realized, and efficiency improvements will take time. Currently, in the AI API business, a router service provider coordinates multiple models or automatically routes models based on cost. This is a good service in itself, but service providers often use inferior services to boost profits. For example, a user may call GPT5 10 times, but the router service provider, in order to increase profits, may give you GPT5 6 times and a lower-cost model 4 times. Of course, ZKML/ZKGPT is not the only solution to this problem. Allora's multi-model coordination network @AlloraNetwork may not have addressed this aspect yet, but I believe they could consider adding this layer of computational verification services. The value of verifiable computation is still on the horizon.